Insect Identification in the Wild: The AMI Dataset
~2.5 million insect images from citizen science platforms (AMI-GBIF) and 2,893 annotated images from global automated camera traps (AMI-Traps) for automated insect monitoring.
- Size
- 14.4 GB
- Format
- ZIP
- License
- MIT
- Access
- Open Access
- Source
- Zenodo (record 11358689)
Your edits to the notebook only affect your own copy in the Hub. Changed it and want the original back? Restore it here.
Challenge
Automated insect monitoring is only as fair as its training labels. In the AMI dataset, which taxa are over- or under-represented, how coarse are the identifications, and where on the globe were the images actually collected?
- Where to start
- Read the annotation tables from inside ami_dataset.zip without extracting. For taxonomic balance, load ami_traps/insect_crops/fgrained_labels.json (one row per crop with label, taxon_rank, region) and count crops per label, per taxon_rank and per region. For geography, read only speciesKey, decimalLatitude and decimalLongitude (usecols) from ami_gbif/fine-grained_classification/metadata/04_ami-gbif_fine-grained_all_val.csv and join taxonomy_map.csv on speciesKey to attach family names.
- What to share
- A ranked bar chart of the most-represented taxa (with the singleton/long-tail count called out), a breakdown of labels by identification rank and region, and an offline world map of the GBIF collection points showing the sampling gaps -- plus a short comment on which regions and rare taxa are most under-sampled.
💬 Discuss this dataset, ask questions and share your results in its discussion thread. Challenges are open-ended — there's no single right answer.